We provide the prompt for generating sequence-based algorithm problems in prompt.py (part of the prompts) and use generation.py for problem generation.
In the rolling folder, we include the code for rolling and sandbox environment testing.
For the training phase, we primarily use the InternTrainer and Verl frameworks, both of which are publicly available LLM training frameworks.
If our paper is accepted, we will release more detailed usage instructions and example data on GitHub.